CT-based delta-radiomics nomogram to predict pathological complete response after neoadjuvant chemoradiotherapy in esophageal squamous cell carcinoma patients

Abstract

Background: This study developed a nomogram model using CT-based delta-radiomics features and clinical factors to predict pathological complete response (pCR) in esophageal squamous cell carcinoma (ESCC) patients receiving neoadjuvant chemoradiotherapy (nCRT).

Methods: The study retrospectively analyzed 232 ESCC patients who underwent pretreatment and post-treatment CT scans. Patients were divided into training (n = 186) and validation (n = 46) sets through fivefold cross-validation. 837 radiomics features were extracted from regions of interest (ROIs) delineations on CT images before and after nCRT to calculate delta values. The LASSO algorithm selected delta-radiomics features (DRF) based on classification performance. Logistic regression constructed a nomogram incorporating DRFs and clinical factors. Receiver operating characteristic (ROC) and area under the curve (AUC) analyses evaluated nomogram performance for predicting pCR.

Results: No significant differences existed between the training and validation datasets. The 4-feature delta-radiomics signature (DRS) demonstrated good predictive accuracy for pCR, with α-binormal-based and empirical AUCs of 0.871 and 0.869. T-stage (p = 0.001) and differentiation degree (p = 0.018) were independent predictors of pCR. The nomogram combined the DRS and clinical factors improved the classification performance in the training dataset (AUC_αbin = 0.933 and AUC_emp = 0.941). The validation set showed similar performance with AUCs of 0.958 and 0.962.

Conclusions: The CT-based delta-radiomics nomogram model with clinical factors provided high predictive accuracy for pCR in ESCC patients after nCRT.

Keywords: Computed tomography; Delta-radiomics; Esophageal squamous cell carcinoma; Neoadjuvant chemoradiotherapy; Pathological complete response.

Fig. 1

Features selection through LASSO with a binary regression model. A The LASSO coefficient profile plot was produced against the log lambda sequence. B Tuning parameter (log lambda) selection in the LASSO via minimum criteria. AUC area under the curve

Fig. 2

The performances of the developed delta-radiomics signature. A Receiver operating characteristics (ROC) curves. B Precision-recall curve (PRC)

Fig. 3

Nomograms developed in this study using the training dataset

Fig. 4

The performance of the developed nomogram. A Receiver operating characteristics (ROC) curves. B Precision-recall curve (PRC)

Fig. 5

Calibration curve of the nomogram shows as a red line

Fig. 6

The decision of the delta-radiomics signature, nomogram and two extreme curves were plotted based on the validation dataset. The figure illustrated that the utilize of nomogram to predict pCR probability has a greater benefit that the delta-radiomics signature